High-precision pest and disease detection in greenhouses using the novel IM-AlexNet framework.

Abstract

China is the largest producer of greenhouse vegetables, but the closed environment fosters high pest and disease incidence. This study proposes an improved AlexNet (IM-AlexNet) model incorporating ReLU6, batch normalization, and GoogleNet Inception-v3 to enhance pest and disease identification. Experimental results show that the IM-AlexNet model is better than the traditional model in indicators such as Precision, Recall, F1, and MAP. Specifically, its MAP value is 88.91%, which is 10.77, 8.6, and 5.14% higher than the AlexNet, CNN, and YOLO-v7 models, which shows stronger generalization capabilities under small sample conditions. It demonstrates strong generalization, reduced missed detection, and improved target recognition in complex backgrounds. This model offers a valuable tool for greenhouse vegetable growers to monitor pests and diseases intelligently, reduce pesticide use, and improve environmental sustainability. The findings provide a foundation for further research in agricultural pest management.